Foresight

Most of the platform answers questions about what did happen. Foresight answers questions about what could happen.

Think of it as a research assistant that scans your data, the web (if you allow it), pinned sources, and your project's history to surface things worth knowing before you make a call.

What's on the Foresight screen

Sidebar → Foresight. Composer-first layout:

• Header card "What do you want to research?" with description: "Ask a forward-looking strategic question. Returns a headline, key assumptions, action cards, and the data gaps the agent noticed."
• Settings button (top-right of the card) — opens the project-level Foresight settings.
• Composer textarea — placeholder like "e.g. What's the biggest risk to our Q2 growth plan?"
• Source-scope pill — 🌐 Open web · pick sources (click to add specific URLs / domains to bias the scan).
• Skip web checkbox — "Skip web — internal data only" restricts the query to your sources and pinned context.
• Mode indicator — text on the right of the controls reads e.g. "Open web + internal projections" and updates as you toggle.
• Run button.

Below the composer, a ✨ Suggested research topics section with a Regenerate action and a grid of cards. Each card is tagged with one of the four modes:

• 🔴 RISK — "What could go wrong with X?"
• 🟢 OPPORTUNITY — "Where could we grow that we aren't?"
• 🟠 TRADEOFF — "What are the side-effects of X?"
• 🔵 SCENARIO — "What if X happened?"

The platform classifies each query into one of these modes and colour-tags the result.

What you get back

Each foresight result has:

• A headline — the most important takeaway in one line.
• A mode tag — risk, opportunity, tradeoff, scenario or monitor.
• A sources scanned count — how much the platform looked at to produce this.
• The body — structured findings, broken out by sub-theme.
• A cite list — every URL, document, and internal source it drew on.

Project-level Foresight settings

Some inputs are set once at the project level (Settings → Foresight) and apply to every query:

• Company profile — your company name, description, market. Helps the platform tailor results.
• Trusted domains — sites the platform should weight heavily (e.g. your industry's analyst firm).
• Excluded domains — sites to ignore (e.g. low-quality aggregators).
• Bookmark URLs — sources to always include in scans.

Why this is separate from Reasoning

Reasoning is best for "what happened?" questions — they have a defensible answer in your data. Foresight is best for "what might?" questions — they require external scanning, scenario modelling, and explicit acknowledgement that the future isn't in your CSV.

Keeping them separate means each one is honest about its evidence base.

Tips

• Pin your sources. A query with two pinned URLs is dramatically better than a free-form web scan.
• Use trusted/excluded domains. It compounds over months.
• Tag the mode in your phrasing. "What risks should we watch for around X" steers cleanly toward Risk mode.
• Don't expect predictions. Foresight surfaces what to think about — not "the answer is yes."